Difference of 4 Life Distribution Groups & Relationship(custom) in Fit Life by X

IkjuShin

Hi everyone,

I got several questions from some Seminconductor Reliability Engineering team.

Among of them, I want to ask three questions;

 

1. There are 4 major groups of parametric distributions in 'Life Distribution'
  -Basic Failure Time Distributions

  -TH(Threshold) Distributions

  -DS(Defective subpopulation) distributions

  -ZI(Zero-inflated) distributions

  I have much difficulty to explain the difference of 4 groups.

  Please let me know if you have any good ideas to explain simply and easily to them.

2. We have to identify the relationship between the event and accelerating factor in 'Fit Life by X'

 1) If there is no pre-defined model or they don't know which relationship is best fit,

     How can they find best fit relationship ?

 2) If they use relationship(custom), how can they input proper values in below blanks ?

   

3. Usual scatterplot is like below in 'Fit Life by X'

  

  If their data are measured or tested in two or more batches or groups,

   Could we plot overlaid scatter plot  grouped by batches or groups ?

 

Thanks

Ikju

  

Mark_Bailey

1. The reliability life distribution models are derived from theoretical models of failure. They generally require two parameters: location and scale or alternatively scale and shape and assume that life begins at zero. This parameterization is insufficient for some cases. The threshold distribution models use a third parameter for cases where the distribution is not anchored at the origin. The threshold represent a non-zero origin. It might be positive (failure-free period of life) or it might be negative (life ended before manufacturing is complete). The zero-inflated models account for the cases that exhibit units with no life ("infant death" or failures during burn-in period). The opposite is the defective sub-population models that account for units with infinite life (failure mode impossible for all units).

 

2.1. The plots are intended to guide the choice of a model in the absence of a theoretical model or prior knowledge of the failure mode. This approach is merely fitting data but it can lead to useful models anyway. The user interface supports a variety of models to test and compare. It is helpful to search the technical and scientific literature for articles about the reliability and failure modes of similar units

 

2.2. See Help > Books > Reliability and Survival > Fit Life by X chapter. There is a section devoted to using custom relationships.

 

3. I do not think that is possible. The batches, though, represent random error anyway.

 

I highly recommend the reference textbook, "Statistical Methods for Reliabilty Data" by Meeker and Escobar. Bill Meeker has been a valued collaborator with JMP Development for the creation of the reliability platforms.